Digital technology has become an increasingly important part of English language education, particularly in higher education contexts where learners engage with English through web-based platforms, mobile applications, online resources, and other multimedia-supported tools. In such environments, English learning is no longer shaped solely by teacher-led classroom instruction but also by learners’ independent use of digital resources for accessing input, preparing output, and managing their own learning processes. As a result, online English education can be understood not simply as a change in delivery mode, but as a broader shift toward digitally mediated and multimedia-supported learning environments in which learner autonomy, self-regulation, and technological support are closely intertwined (Godwin-Jones, 2023; Lo, 2023; Niu et al., 2022).

Recent work has also shown that digital EFL learning is associated not only with access to learning technologies but also with learners’ digital competence and their ability to cope with technostress in digitally intensive learning contexts (Niu et al., 2022). In addition, research on ESL online classrooms in higher education suggests that online English teaching requires substantial pedagogical adjustment in areas such as interaction, task design, and learner support, further highlighting the importance of examining English learning within fully online instructional settings (Lo, 2023). From a broader perspective, digital and AI-mediated environments have expanded the spaces in which language learning occurs, making learner autonomy, self-regulation, and interaction with digital resources increasingly central to language learning experiences (Godwin-Jones, 2023; Huang et al., 2022). These studies suggest that online English learning should be viewed as part of a broader digital learning ecology rather than as a simple transfer of conventional classroom instruction to online platforms.

Within the Korean university context, this shift has also become visible in courses that incorporate digital platforms and technology-mediated interaction into English teaching. For example, Kim (2022) examined the use of Padlet in non-face-to-face liberal arts English classes and reported that the platform functioned as a useful web-based tool for participation, communication, and collaborative sharing in online instruction. Related local studies have also suggested that learner experiences in online general English courses may involve interaction, instructional support, affective responses, and performance-related outcomes beyond language achievement alone (Chong, 2023, 2025). Taken together, these studies indicate that digital and multimedia-supported English learning environments provide not only content access but also opportunities for learner engagement, collaboration, and self-directed participation.

Previous studies have demonstrated that English learning through media and technology can take multiple pedagogical forms. Digital and multimedia-supported learning environments may provide learners with additional opportunities for input processing, output practice, interaction, feedback, and reflection beyond conventional classroom instruction (Godwin-Jones, 2023; Huang et al., 2022; Jeon et al., 2023). In particular, chatbot-supported and speech-recognition-based learning environments have been examined as ways to support language practice, learner engagement, and oral performance, although their pedagogical value depends on task design, learner interaction, and the quality of technological support (Huang et al., 2022; Jeon et al., 2023).

Within this broader line of research, media-based materials and technology-supported output activities have also been explored in Korean EFL contexts. For example, earlier studies examined English TV news-based reading aloud and integrated tasks combining writing, reading aloud, and voice recording, showing that such activities could support learner involvement, oral practice, and confidence development (Chong, 2021, 2022). These studies provide a local foundation for examining how media- and technology-supported English learning may operate in Korean university contexts. However, they focused mainly on specific classroom activities rather than on students’ broader perceptions of AI-supported learning across fully online general English courses.

More recent studies have moved beyond media-based classroom tasks toward broader questions of AI and digital support in English learning. In particular, AI-assisted writing, translation, feedback, chatbot interaction, and selfdirected learning support have become increasingly important areas of inquiry in EFL research. This shift provides the basis for examining how students perceive AI-related tools not only as task-specific aids but also as part of their wider online English learning practices.

The emergence of generative AI has added a new dimension to digitally supported English learning. Recent reviews suggest that ChatGPT and other AI-based tools have been used across multiple areas of language education, including writing, speaking, feedback, translation, brainstorming, and individualized learning support (Kohnke et al., 2023; Li et al., 2024; Lo et al., 2024). At the same time, these studies also point to unresolved issues such as reliability, ethical use, learner dependence, pedagogical design, and the need for teacher and learner digital competence. This suggests that AI-supported learning in English education is not limited to a single task type but is developing as a broader pedagogical area across different language skills and learning contexts.

Previous studies have especially emphasized the use of AI for writing-related support. Bok and Cho (2023) noted that learners responded favorably to ChatGPT when using it for writing revision, while still recognizing limitations related to trust, explanation quality, and authorship. Mun (2024) similarly showed that AI-assisted writing feedback may support students’ revision processes and reduce certain forms of language error, although students’ reflections also revealed caution toward overdependence. More broadly, Nguyen et al. (2024) examined ChatGPT as a tool for self-learning English among EFL learners and showed that ChatGPT can be positioned as a resource for self-directed English learning rather than only as a task-specific writing aid. These findings suggest that AI is pedagogically meaningful not simply because it automates feedback but because it changes how learners approach planning, revising, and monitoring their own language production.

Research has also begun to examine AI use beyond writing revision alone. Koh (2023) analyzed ChatGPT’s Korean-to-English translation in relation to film-based language material and argued that, with appropriate prompting, ChatGPT could function as a useful language-learning and translation support tool for EFL learners. Although the study centered on translation, it is relevant to the present research because it illustrates how AI may operate within media-rich English learning contexts rather than only within isolated writing exercises. In addition, research on chatbot- and speech-recognition-based language learning has shown that AI-mediated interaction can support speaking practice, confidence, engagement, and pronunciation-related learning, although such effects depend on task design and the quality of learner interaction with the tool (Du & Daniel, 2024; Huang et al., 2022; Jeon et al., 2023).

In addition to task-specific uses of AI, recent studies suggest that the educational value of AI depends on how learners incorporate such tools into their own learning processes. AI tools may provide immediate support for grammar checking, vocabulary selection, translation, idea generation, and revision, but these forms of assistance do not automatically lead to stronger confidence or more active language use. From this perspective, AI-supported learning should be examined in relation to learner engagement, self-regulation, and strategy use, particularly in online courses where students are expected to manage much of their learning outside real-time classroom interaction (Kohnke et al., 2023; Nguyen et al., 2024; Xu et al., 2024).

Taken together, these studies indicate that AI-supported learning in English education should be viewed not merely as automated assistance, but as a developing area of pedagogical support whose value depends on how learners and instructors incorporate it into actual learning practices.

Although AI-supported tools may provide useful assistance, technology alone does not determine learners’ affective and behavioral responses. In second and foreign language education, anxiety has long been recognized as a key factor that shapes classroom participation and performance. Horwitz et al. (1986) conceptualized foreign language classroom anxiety as a distinct, situation-specific form of anxiety associated with communication apprehension, test anxiety, and fear of negative evaluation. Their framework remains foundational for understanding why learners may hesitate, avoid participation, or feel psychologically burdened even when they possess relevant linguistic knowledge.

Alongside anxiety, learner engagement, self-efficacy, and strategic behavior have also been identified as important in online English learning. For example, Wu (2023) found that online learning self-efficacy, informal digital learning of English, and student engagement in online classes were positively interconnected, with social presence playing a mediating role in this relationship. Similarly, Teng and Wu (2024) reported that self-efficacy beliefs predicted learners’ use of metacognitive strategies, which in turn predicted language learning motivation and perceived online English learning progress. In a recent study of EFL students in smart classrooms, Xu et al. (2024) further demonstrated that self-regulated learning strategies, self-efficacy, and learning engagement are closely interconnected in technologyrich EFL learning environments. Their findings support the view that learners’ confidence and engagement in digitally mediated English learning are shaped not merely by access to technology but by how learners regulate, evaluate, and sustain their own learning processes. Together, these studies suggest that progress and confidence in online English learning depend not only on access to digital tools but also on how actively and strategically learners regulate their own study processes. In a related local study on online general English instruction, students’ emotional responses, including anxiety and confidence, were also found to be associated with engagement and performance-related outcomes across different online instructional formats (Chong, 2025).

This perspective is highly relevant to the present study because students’ perceptions of AI helpfulness may not directly translate into stronger confidence unless such support is accompanied by meaningful engagement and strategic use. In other words, performance confidence in online English learning may depend less on tool access itself than on how learners incorporate digital support into sustained, purposeful language practice.

The studies reviewed above provide important insights into media-based English learning, technology-supported classroom activities, and recent uses of AI in English education. Previous work has shown the pedagogical value of media-based input and output activities, such as English TV news-based reading aloud and voice recording, as well as web-based tools such as Padlet and AI-assisted writing support (Bok & Cho, 2023; Chong, 2021, 2022; Kim, 2022; Mun, 2024). Broader studies have also examined chatbot-supported language learning, speech-recognition-based practice, AI-assisted feedback, and ChatGPT use in language education (Du & Daniel, 2024; Huang et al., 2022; Jeon et al., 2023; Kohnke et al., 2023; Li et al., 2024; Lo et al., 2024). However, much of the existing literature has focused either on specific activity designs or on AI use within relatively narrow task domains such as writing revision, translation, chatbot interaction, or speaking practice.

At the same time, broader studies on online and digital English learning have emphasized the importance of learner engagement, self-efficacy, digital competence, and affective response in shaping learning experiences (Niu et al., 2022; Teng & Wu, 2024; Wu, 2023; Xu et al., 2024). However, relatively less attention has been paid to how perceived AI helpfulness is related to engagement/strategy, anxiety in using English, and performance confidence in post-course survey contexts following fully online general English instruction. This issue is particularly relevant because students in such courses may use AI-related tools not as part of a controlled instructional treatment but as optional support resources within their regular online learning practices.

In particular, there remains a need to examine AI-supported learning not as the effect of a single instructional intervention but as part of a wider web-based multimedia learning environment in which students interpret, use, and respond to digital support in different ways. The present study addresses this gap by analyzing post-course survey data from fully online general English instruction and by examining how perceived AI helpfulness is positioned alongside learner engagement, anxiety, and confidence.

This study employed an exploratory survey-based design to examine university students’ perceptions of AIsupported English learning in fully online general English instruction. The primary focus of the study was on the relationships among perceived AI helpfulness, engagement/strategy, anxiety in using English, and performance confidence. Because the study was conducted after the completion of regular course instruction, it did not involve an experimental intervention or random assignment. Instead, it explored how students perceived and used AI-related support tools within their actual online English learning experience. Selected learner background variables were examined for descriptive and exploratory purposes, and open-ended responses were analyzed to provide supplementary contextual information for interpreting the quantitative findings.

The data were drawn from a voluntary online post-course survey administered after the completion of fully online general English instruction during the first semester of 2025. The study was conducted at a private university located in Gimhae, South Korea. The participants were undergraduate students enrolled in liberal arts English courses. At the end of the semester, students were invited to complete an online survey about their English learning experience, use of AI-related tools, and perceptions of AI-supported learning. Participation was voluntary, and a total of 51 valid responses were included in the final analysis.

The participants were drawn from three fully online general English course types: an English news and expression decoding course, a practical English course, and a speaking-oriented course. Although the specific topics and assignments differed across courses, all courses were delivered asynchronously through the university’s LMS and required students to watch pre-recorded lectures, study English materials, complete LMS-based assignments, and prepare written or spoken English output. The common instructional goals were to support students’ comprehension of English materials, development of useful vocabulary and expressions, preparation of English output, and confidence in English-related performance tasks. These instructional features provided a relevant context for examining how students used AI-related and web-based tools to support English learning.

AI-related and web-based tools were not implemented as a formal instructional intervention in these courses. Rather, they were available as optional support resources that students could use while completing regular online learning tasks. In a typical learning sequence, students first watched pre-recorded lectures and studied the assigned English materials through the LMS. They then completed LMS-based assignments that required them to understand English input, check vocabulary and expressions, organize ideas, and prepare written or spoken output. During these preparation and revision stages, students could use tools such as machine translation, ChatGPT, online dictionaries, or search-based resources to check meanings, revise expressions, generate ideas, or rehearse output. Students were advised to use such tools as learning support rather than as a substitute for their own work. In particular, they were expected to review, adapt, and understand AI-generated or translated output before submitting assignments. Thus, the present study examined students’ post-course perceptions of AI-supported learning as it occurred within their regular online learning practices, rather than evaluating the effect of a controlled AI-based instructional treatment.

Before analysis, identifying information was removed from the dataset. Responses to the grade-level item were cleaned because they had been entered in free-response format. Valid responses corresponding to years 1 through 4 were retained, and clearly invalid entries were treated as missing. No substantial missingness was observed in the main scale variables used in the quantitative analyses. For the multiple-response items on AI tool use and perceived areas of AI benefit, frequencies and percentages were calculated using the total number of valid respondents as the denominator.

Descriptive statistics were calculated for the major variables, including perceived AI helpfulness, engagement/strategy, anxiety in using English, and performance confidence. Frequencies and percentages were also calculated for AI-related tool use and perceived areas of AI benefit. Because the AI tool use and perceived benefit items allowed multiple responses, percentages were calculated based on the total number of participants rather than the total number of responses.

Internal consistency was assessed using Cronbach’s alpha for the major scales and subscales. Exploratory factor analysis was conducted to examine the structure of the 12 attitude items and the five AI helpfulness items. The suitability of the data for factor analysis was assessed using the Kaiser-Meyer-Olkin (KMO) measure and Bartlett’s test of sphericity. Based on the factor-analytic results, subscale mean scores were calculated and used in subsequent analyses.

Selected group comparisons were conducted for exploratory purposes according to English-speaking-country experience and standardized English test experience. These comparisons were interpreted cautiously because of the small sample size and unequal subgroup distribution.

A multiple regression analysis was conducted to examine predictors of performance confidence. Performance confidence was entered as the dependent variable, and perceived AI helpfulness, engagement/strategy, anxiety in using English, English-speaking-country experience, and standardized English test experience were entered as predictors. Before interpreting the model, the major assumptions of multiple regression were examined. Linearity was inspected through scatterplots and residual plots. Independence of errors was checked using the Durbin-Watson statistic. The normality and homoscedasticity of residuals were examined through residual plots and normal probability plots. Multicollinearity was assessed using tolerance and variance inflation factor values. No serious violation was observed in the diagnostic checks, and the VIF values did not indicate problematic multicollinearity.

Open-ended responses were reviewed using a brief qualitative content analysis procedure. The responses were first read repeatedly to identify recurring themes related to perceived instructional benefits and areas for improvement. Similar responses were then grouped into broad categories, such as speaking and presentation practice, materialsbased learning, career-related English, workload, assessment, and limitations of non-face-to-face instruction. Representative comments were selected to illustrate the major response patterns. The qualitative results were not treated as independent evidence of learning outcomes but were used to provide supplementary context for interpreting the quantitative findings.

The study was based on de-identified survey responses collected from students after the completion of instruction. Personal identifiers were removed during data preparation, and the analysis focused on aggregated patterns rather than individual cases.

The final sample consisted of 51 students who completed a voluntary online post-course survey after fully online general English instruction. As shown in Table 2, the participants were enrolled in three course types: English news, practical English, and speaking-oriented English. Most students were in their second or fourth year. Six students (11.8%) reported prior experience in an English-speaking country, and 24 students (47.1%) reported prior experience taking a standardized English test. These background variables were included mainly to describe the sample and to support limited exploratory interpretation rather than to function as focal explanatory variables.

Students reported using multiple AI-related and web-based learning tools rather than relying on a single resource. Because multiple responses were allowed, the percentages were calculated based on the total number of respondents. The most frequently reported tools were Papago (68.6%), ChatGPT (62.7%), and online dictionaries (54.9%), followed by Google-based resources (31.4%). This pattern suggests that students’ English learning in the present instructional context was supported by a mixed digital environment combining machine translation, generative AI, and reference tools.

The descriptive statistics in Table 3 showed a clear contrast among the major variables. On the 5-point scales, engagement/strategy showed a mid-range mean (M = 3.34, SD = 0.86), anxiety in using English was relatively high (M = 3.49, SD = 1.05), and performance confidence was relatively low (M = 2.84, SD = 0.84). By contrast, perceived AI helpfulness was above the midpoint (M = 3.55, SD = 0.74).

This pattern is noteworthy because it suggests that students generally regarded AI as somewhat helpful, but such perceived usefulness did not coincide with equally strong confidence in performing English tasks. In other words, practical support and psychological readiness did not appear to develop at the same level. This finding supports the need to examine AI-supported learning not only in terms of convenience or usefulness but also in relation to learner attitudes and affective variables.

The reliability analysis showed satisfactory to high internal consistency for the major scales. Table 4 summarizes the reliability coefficients for the major scales. Cronbach’s alpha was .844 for the 12 attitude items, .856 for engagement/strategy, .886 for anxiety in using English, .903 for confidence, and .916 for perceived AI helpfulness. These values indicate that the measures were sufficiently reliable for exploratory analysis. Because the anxiety factor consisted of two items, its reliability coefficient was interpreted cautiously and is revisited as a limitation in the Conclusion.

Exploratory factor analysis also supported the structural validity of the key measures. For the 12 attitude items, the KMO value was .737, and Bartlett’s test of sphericity was significant (p < .001), supporting the suitability of the data for factor analysis. A three-factor solution was adopted, consisting of engagement/strategy, anxiety in using English, and performance confidence. The engagement/strategy factor was defined primarily by items related to practicing input and output outside class and knowing how to improve English output. The anxiety factor reflected tension and worry in using English, while the performance confidence factor was represented by self-confidence in writing, speaking, presentation, reading aloud, and pronunciation. The factor loadings for the 12 attitude items are presented in Table 5.

For the five perceived AI helpfulness items, the KMO value was .819, and Bartlett’s test was significant (p < .001). These results supported the suitability of the data for factor analysis.

As shown in Table 6, all five perceived AI helpfulness items loaded on a single factor, with factor loadings ranging from .667 to .942. This result suggests that students tended to perceive AI helpfulness as a general support construct across multiple English learning areas rather than as assistance limited to one specific task.

Exploratory comparisons were conducted according to English-speaking-country experience and standardized English test experience. Because only six students reported prior experience in an English-speaking country (n = 6), the results related to this variable were interpreted only as descriptive and exploratory patterns rather than as evidence of reliable subgroup differences. In the regression model, English-speaking-country experience was not a significant predictor of performance confidence (β = .101, p = .312). Therefore, subgroup patterns based on English-speakingcountry experience were not emphasized in the interpretation of the study. Standardized English test experience was retained as a background variable because it may reflect students’ prior exposure to test-oriented English learning and was examined further in the regression model.

A multiple regression analysis was conducted to identify predictors of performance confidence. The model included perceived AI helpfulness, engagement/strategy, anxiety in using English, English-speaking-country experience, and standardized English test experience. The model showed moderate explanatory power (R² = .555, adjusted R² = .470).

As shown in Table 7, engagement/strategy was the strongest positive predictor of performance confidence (b = 0.590, standardized β = .604, p < .001). In contrast, perceived AI helpfulness did not significantly predict performance confidence in the final model (b = 0.179, β = .157, p = .216). Anxiety in using English showed a negative but nonsignificant tendency (p = .086). Standardized English test experience negatively predicted performance confidence (b = -0.487, β = -.294, p = .008), whereas English-speaking-country experience was not a significant predictor.

Diagnostic checks indicated that the regression model was appropriate for exploratory interpretation. In particular, the VIF values did not indicate problematic multicollinearity, and the residual diagnostics did not show serious violations of the regression assumptions.

This set of findings is central to the interpretation of the study. Although students generally perceived AI as helpful, such perceived helpfulness alone did not appear to function as a sufficient direct predictor of confidence in English performance. Instead, performance confidence was more strongly associated with active and strategic learner engagement. This suggests that AI may be most educationally meaningful when it is incorporated into purposeful practice, preparation, and self-directed learning routines rather than being treated as a stand-alone source of support.

This set of findings is central to the interpretation of the study. Although students generally perceived AI as helpful, such perceived helpfulness alone did not appear to function as a sufficient direct predictor of confidence in English performance. Instead, performance confidence was more strongly associated with active and strategic learner engagement. This suggests that AI may be most educationally meaningful when it is incorporated into purposeful practice, preparation, and self-directed learning routines rather than being treated as a stand-alone source of support.

The negative association between standardized English test experience and performance confidence is also noteworthy. One possible interpretation is that students with test-oriented English learning experience may apply stricter evaluative standards to their own performance, especially in speaking- or presentation-related contexts. In this sense, confidence in English performance may be shaped not only by perceived support but also by the learner’s prior orientation to assessment and productive language use.

This finding suggests that the educational value of AI in fully online English instruction may lie less in the tool itself than in how learners incorporate it into active and strategic learning behaviors. In this respect, the present study extends previous AI-related studies that focused mainly on task-specific usefulness by emphasizing the mediating role of learner engagement and strategy use.

These findings both align with and extend previous studies on AI-supported English learning. The students’ positive perceptions of AI, especially for grammar, writing, and vocabulary support, are consistent with previous studies showing that ChatGPT or AI-assisted tools can support writing revision, feedback, and language-form checking (Bok & Cho, 2023; Lo et al., 2024; Mun, 2024). However, the present regression results suggest that perceived AI helpfulness alone was not sufficient to explain confidence in English performance. Rather, engagement/strategy was the strongest predictor, indicating that AI may be educationally meaningful when learners incorporate it into active preparation, practice, and self-directed learning. This interpretation is consistent with online EFL learning research emphasizing the role of self-regulation, engagement, and self-efficacy in shaping learners’ perceived progress and confidence (Teng & Wu, 2024; Wu, 2023; Xu et al., 2024).

Students’ responses regarding the areas in which AI was helpful were collected through a multiple-response item. Therefore, the percentages reported here were calculated based on the total number of participants (N = 51). The most frequently selected areas were grammar (52.9%), writing (49.0%), and vocabulary (49.0%), followed by pronunciation (39.2%) and presentation (31.4%). Lower selection rates were observed for speaking (23.5%), reading aloud (21.6%), and listening (17.6%).

This distribution suggests that students mainly perceived AI as beneficial for language-form support and preparation rather than for immediate real-time performance. That is, AI appears to have been especially useful when learners needed help with checking language, organizing output, or preparing for tasks, but less useful when they faced the more immediate demands of oral production. This interpretation is consistent with the earlier finding that perceived AI helpfulness was not itself a strong direct predictor of performance confidence.

Open-ended responses were optional, and not all students provided written comments. Therefore, these responses were used as supplementary qualitative context for interpreting students’ broader online learning experiences rather than as direct evidence of AI-related learning effects. Although these comments did not always refer directly to AI tool use, they help clarify the kinds of learning tasks in which students could use AI-related and web-based tools for support, including understanding English materials, preparing presentations, practicing speaking, and completing career-related English assignments. Among the written responses, students referred to useful aspects of the course such as materials-based learning, opportunities to prepare English output, and career-related English. For example, one student noted that the course provided regular opportunities to practice speaking English, which was difficult to do outside class. Another student identified “speaking English directly and giving presentations” as a useful experience. One response indicated that it was helpful to encounter international news in English, while another stated that repeated learning through English news helped them practice authentic expressions and pronunciation. Careerrelated English was also mentioned, with one student noting that learning how to write a resume in English was particularly useful. These responses suggest that students valued structured preparation and meaningful outputoriented tasks, which may help explain why engagement and strategy use played an important role in the quantitative results.

Taken together, the findings of the present study suggest that the pedagogical value of AI in fully online general English instruction lies not simply in the availability of the tool itself but in how learners are guided to use it. The results indicate that AI is already embedded in students’ study practices and is generally perceived as helpful. However, stronger performance confidence appears to depend more on engagement, strategy use, and active participation than on perceived AI helpfulness alone.

This implies that instructors should move beyond simply allowing students to use AI tools. Instead, they should provide clearer guidance on how such tools can be used for planning, rehearsal, self-monitoring, revision, and reflection. In addition, instructional design should continue to promote structured preparation and meaningful output practice because students’ comments suggest that they valued materials-based learning, presentation preparation, and opportunities to organize English output. In this sense, AI may be most effective when it is integrated into structured online learning routines that encourage purposeful engagement rather than passive dependence.